1

Crustal and Uppermost Mantle Shear Velocity Structure Adjacent to the 1

Juan de Fuca Ridge from Ambient Seismic Noise 2

Ye Tian, Weisen Shen, and Michael H. Ritzwoller 3

Center for Imaging the Earth’s Interior, Department of Physics, University of Colorado at 4

Boulder, Boulder, CO 80309, USA (Ye.Tian@colorado.edu) 5

6

Key points 7

Rayleigh wave velocities vary with crustal age near the Juan de Fuca ridge. 8

Mantle Vs increases with crustal age faster than conductive cooling. 9

A shallow low velocity zone near the ridge implies less than 1% melt fraction. 10

11

Abstract 12

Based on six months of OBS data from the Cascadia Initiative experiment near the Juan 13

de Fuca Ridge, we obtain Rayleigh wave group and phase speed curves from 6 sec to 14

about 20 sec period from ambient noise cross-correlations among all station-pairs. We 15

confirm the hypothesis that the dispersion data can be fit by a simple age-dependent 16

formula, which we invert using a Bayesian Monte Carlo formalism for an age-dependent 17

shear wave speed model of the crust and uppermost mantle between crustal ages of 0.5 18

Ma and 3.5 Ma. Igneous crustal structure is age-invariant with a thickness of 7 km, water 19

2

depth varies in a prescribed way, and sedimentary thickness and mantle shear wave 20

speeds are found to increase systematically with crustal age. The mantle model possesses 21

a shallow low shear velocity zone (LVZ) with a velocity minimum at about 20 km depth 22

at 0.5 Ma with lithosphere thickening monotonically with age. Minimum mantle shear 23

velocities at young ages are lower than predicted from a half-space conductively cooling 24

model (HSCM) and the lithosphere thickens with age faster than the HSCM, providing 25

evidence for non-conductive cooling in the young lithosphere. The shallow LVZ is 26

consistent with expectations for a largely dehydrated depleted (harzburgite) mantle with a 27

small, retained near-ridge partial melt fraction probably less than 1% with melt extending 28

to a lithospheric age of approximately 1 Ma (i.e., ~30 km from the ridge). 29

30

Keywords: ambient noise, seismic inversion, mid-oceanic ridge, lithospheric age, surface 31

waves, lithospheric structure, low-velocity zone, partial melt, Juan de Fuca plate 32

33

3

1. Introduction 34

Seismic information on the early evolution of the oceanic mantle lithosphere near 35

spreading ridges has been derived principally from the MELT and GLIMPSE 36

experiments [e.g., MELT Seismic Team, 1998; Harmon et al., 2009; Yao et al., 2011] near 37

the East Pacific Rise (EPR), a fast spreading ridge with a full spreading rate of about 14 38

cm/yr. The recent deployment of ocean bottom seismographs (OBS) by the Cascadia 39

Initiative on the Juan de Fuca Plate and the open availability of these data provide the 40

opportunity to characterize the mantle lithosphere near a slower spreading ridge (~6 41

cm/yr) and ultimately to extend analyses to the entire plate. Harmon et al. [2007] and Yao 42

et al. [2011] showed that short period Rayleigh waves and the first higher mode can be 43

observed using cross-correlations of ambient noise recorded on OBS installed near the 44

EPR. They used these waves to constrain shear wave speeds in the oceanic crust and 45

uppermost mantle. Here, we analyze cross-correlations of the first six-months of ambient 46

noise recorded by OBS installed near the Juan de Fuca ridge in order to determine shear 47

wave speeds in the crust and uppermost mantle in the young Juan de Fuca plate to an age 48

of about 3.5 Ma (i.e., to distances up to about 100 km from the ridge crest). 49

Our goal is to reveal the age dependent structure of the shallow oceanic lithosphere in the 50

young Juan de Fuca plate in order to illuminate the physical processes at work there. In 51

particular, we are interested in modeling the accumulation of sediments and the variation 52

of shear wave speeds in the uppermost mantle to a depth of about 60 km. Like Harmon et 53

al. [2009] for the region near the EPR, we compare the estimated mantle shear wave 54

speeds with those predicted from a conductively cooling half-space to test for the 55

presence of non-conducting cooling processes (e.g., convection, fluid advection, lateral 56

4

heat flux). In addition, we compare with the more sophisticated physical model of Goes 57

et al. [2012] in order to investigate whether dissolved water or interstitial partial melt are 58

present. Goes et al. [2012] argue for a double low velocity zone (LVZ) with a shallow 59

LVZ between about 20 and 50 km depth caused by dry (or damp) partial melting near to 60

the spreading ridge and a deeper LVZ between about 60 and 150 km caused by solid-state 61

anelasticity where low Q values result from dissolved water. Our model, however, 62

extends only to a depth of 60 km and provides no information about a deeper LVZ. 63

2. Methods 64

2.1 Data Processing 65

The Cascadia Initiative (CI) experiment provides the OBS data for this study based on 66

instruments from three different contributors: SIO, LDEO, and WHOI. Because the CI 67

team discovered a (subsequently corrected) timing error that affected the SIO data, we 68

focus attention on the WHOI data near the Juan de Fuca Ridge. This restricts analysis to 69

23 stations. Stations G03A, G30A and J06A are outside of the study area and are, 70

therefore, not used and the vertical channel of station J48A failed during the deployment. 71

We analyze only the long period (1 sps) channel at each station, which eliminates station 72

J61A and restricts our analysis to Rayleigh waves above about 6 sec period. Figure 1a 73

shows the study area and the 18 stations used, 15 of which are located to the east of the 74

Juan de Fuca ridge and provide path coverage up to about 200 kilometers into the Juan de 75

Fuca plate. Approximately six-months of continuous data are available for most of these 76

stations. When we downloaded the data, horizontal components had not yet been rotated 77

5

into the east-west and north-south directions. Therefore, we do not use horizontal data, 78

but restrict analysis to the vertical components (and therefore Rayleigh waves). 79

We computed ambient noise cross-correlations between the vertical components of all 80

stations by applying traditional ambient noise data processing (time domain 81

normalization, frequency domain normalization) to produce the empirical Green’s 82

functions [Bensen et al., 2007]. An example of an empirical Green’s function between 83

stations J47A and J29A (Fig. 1b) is shown in Figure 1c. The Rayleigh waveforms are 84

highly dispersed and display two Airy phases such that the short period phase 85

(representative of the water – sediment waveguide) arrives far after the longer period 86

phase (representative of the igneous crust and uppermost mantle waveguide). Frequency-87

time analysis [e.g., Levshin et al., 2001; Bensen et al., 2007] is applied to the symmetric 88

component (average of positive and negative correlation lags) of each cross-correlation to 89

measure Rayleigh wave group and phase speeds between periods of about 6 and 20 sec. 90

Longer periods require longer time series lengths and may be obtainable as more data 91

become available. An example frequency-time analysis (FTAN) diagram is presented in 92

Figure 1d showing both the Rayleigh wave group and phase speed curves. Rayleigh wave 93

group speeds range from about 1 km/s at the short period end to more than 3.6 km/s at 94

longer periods and phase speeds range from about 1.8 km/s to more than 3.6 km/s. At 95

periods below 6 sec the phase and group speed curves would approach each other 96

asymptotically, but are separate in the observed period band. Harmon et al. [2007] and 97

Yao et al. [2011] observed the first higher mode below 6 sec period, which cannot be 98

observed with the long period data used in our study. Paths that are mainly to the west of 99

the ridge are discarded because they reflect the structure of the Pacific plate and may be 100

6

more affected by the Cobb hotspot (Fig. 1a). Dispersion measurements for paths shorter 101

than three wavelengths are also discarded. As discussed in the following paragraph, data 102

are also selected based on signal-to-noise ratio (SNR) and the agreement of dispersion 103

measurements obtained on the positive and negative lag components of the cross-104

correlations. Finally, a total of 106 inter-station paths are accepted and plotted in Figure 105

1b. 106

As a measure of measurement uncertainty and to search for possible timing errors, we 107

compare phase speed measurements obtained from the positive and negative lag 108

components of the cross-correlations. Not all cross-correlations have arrivals on both lags, 109

but 65 of the 106 inter-station measurements have a signal-to-noise ratio (SNR) greater 110

than 5 on both lags at 14 sec period, which allows for the comparison of inter-lag travel 111

times shown in Figure 2a. We make the assumption that the inter-lag travel time 112

differences are normally distributed and estimate the standard deviation of the entire 113

population to be 0.77 sec from the standard deviation of the travel time differences of the 114

65 inter-station paths. For these 65 inter-station measurements, if the discrepancy 115

between the positive and negative lag phase times (or more accurately, times of outgoing 116

and incoming waves) is less than 1% we average the positive and negative lag cross-117

correlations (forming the symmetric signal) and measure group and phase velocities 118

using the resulting signal. For the remaining inter-station measurements, we use only the 119

lag with the higher SNR and retain the measurement if the SNR on that lag is greater than 120

5. The comparison between phase travel times on the positive and negative lags can also 121

be used to detect timing errors [e.g., Stehly et al., 2007; Lin et al., 2007]. Figure 2b 122

presents the mean difference for each station between the measurements of outgoing and 123

7

incoming phase times at 14 sec period. The 1-σ and 2-σ intervals are computed based on 124

the population standard deviation and the number of measurements for each station. As 125

seen in Figure 2b, the measurement means are all within the 2-σ interval and no station 126

displays an absolute difference in the mean larger than 0.5 sec. This is interpreted as 127

evidence that there is no differential timing error amongst the data that we use in this 128

study, which all come from WHOI. 129

The resulting path coverage (Fig. 1b) is not ideal to produce Rayleigh wave group or 130

phase speed maps using either traditional tomographic methods [e.g., Barmin et al., 2001] 131

or eikonal tomography [Lin et al., 2009]. For this reason, we proceed by testing the 132

hypothesis that Rayleigh wave phase and group speeds depend principally on lithospheric 133

age. At each period, we follow Harmon et al. [2009] and test a velocity-age relationship 134

of the following form: 135

v = c0 + c1 a + c2a (1) 136

where v represents either the observed inter-station Rayleigh wave group or phase 137

velocity, a represents the seafloor age in millions of years (Ma), and c0, c1, and c2 are 138

period dependent unknowns that differ for phase and group speeds and which we attempt 139

to estimate. 140

For each measurement type (phase or group) and each period extending discretely from 6 141

sec to 20 sec, we estimate the three coefficients c0,c1,and c2

. The wave travel time along 142

a path is given by the following path integral, which occurs over a path whose 143

8

dependence on crustal age is prescribed by the lithospheric age model of Mueller et al. 144

[1997] shown in Figure 1b: 145

146

147

To determine the set of best fitting coefficients at each period, we perform a grid search 148

to minimize the total squared misfit: 149

(Sipath

tipath

- vipath )2

i

å , (3) 150

where Sipath,ti

path, and vipath are the inter-station path length, the predicted travel time for a 151

particular choice of c0,c1,and c2

, and the observed wave speed for the ith path, 152

respectively. 153

Figure 3a-b summarizes the resulting estimates of Rayeigh wave phase and group 154

velocity versus lithospheric age at periods of 7, 8, 10, and 15 sec. At short periods, 155

velocities decrease with age because water depth and sedimentary thickness increase. At 156

longer periods, they increase with age because they are sensitive to the cooling mantle. In 157

Figure 3a-b, in order to illustrate the fit to the data we over-plot the estimated velocity-158

age curves with the inter-station observations presented at the average of the lithospheric 159

ages of the two stations. The simple velocity versus age curves given by equation (1) 160

capture the trend in these inter-station group and phase speed measurements, although 161

associating each measurement with a single lithospheric age is not entirely appropriate. 162

An F-test shows that the square root term is only important at periods longer than about 9 163

t path =ds

c0 + c1 a + c2apath

ò . (2)

9

sec, while the linear term is, in general, important at periods below 14 sec. This is 164

expected because the shorter periods are controlled mainly by the linear thickening of the 165

combination of water and sediments, whereas the longer periods are primarily sensitive to 166

mantle thermal structures which change approximately proportionally to the square root 167

of lithospheric age. 168

Fully accurate phase velocity misfit (blue) histograms at 7 and 15 sec period are 169

presented in Figure 3c-d for the age-dependent model, with the standard deviation (std) of 170

misfit of about 1.8% and 0.9%, respectively, and mean misfits less than 0.1%. These 171

values represent a large improvement compared to any age-independent model. For 172

example, the misfit using our estimated phase speed model at 0.5 Ma is presented in 173

Figure 3c-d with the red histograms. The one standard deviation misfit using this model 174

is 5.7% and 1.4% at 7 and 15 sec period, respectively, with mean misfits of -9.7% and 175

3.2%. Because group velocity is a more difficult observable with larger uncertainties than 176

phase velocity, the final misfit is higher but is still substantially better than any age-177

independent model. Our age-dependent model neglects azimuthal anisotropy. However, 178

we did estimate azimuthal anisotropy at all periods and found that the expected bias in 179

isotropic shear wave speed is less than about 0.3% at all periods, which is within 180

estimated uncertainties. 181

In conclusion, the fit to the observations by the Rayleigh wave phase velocity versus age 182

model presented by equation (1) is sufficient to base further interpretation exclusively on 183

the age dependence of the group and phase velocities. Although other spatially dependent 184

variations in Rayleigh wave speeds are expected to exist (and are interesting in their own 185

right), they can be ignored safely in our analysis, which aims to produce an age-186

10

dependent model for the crust and uppermost mantle for the young Juan de Fuca plate. 187

The final result of the data analysis is a set of age-dependent Rayleigh wave phase and 188

group velocity curves such as those at 1 Ma and 3 Ma shown in Figure 4. The error bars 189

are the one standard deviation misfits to the observations given by the estimated age-190

dependent curves such as those shown in Figure 3a-b. 191

2.2 Bayesian Monte Carlo Inversion 192

Examples of the data and uncertainties at 1 Ma and 3 Ma are presented in Figure 4. We 193

are particularly interested in interpreting the age dependence of such curves, which is 194

affected by water depth, sedimentary thickness, crustal thickness, uppermost mantle shear 195

wave speeds, and anelasticity. The shear velocity model we produce is actually a Vsv 196

model because it derives exclusively from Rayleigh waves. 197

2.2.1 Parameterization and constraints 198

At each age, our model is composed of four layers. (1) The top layer is water with a depth 199

that is averaged over the study area as a function of crustal age using a global bathymetry 200

data base [Amante and Eakins, 2009] in which Vs is set 0 km/sec and Vp is 1.45 km/sec. 201

(2) The second layer comprises the sediments with a constant shear wave speed of 1 202

km/sec [Sun, 2000] but with a thickness that varies with age. (3) The igneous crust 203

underlies the sediments and is parameterized by four cubic B-splines. (4) Finally, there is 204

an uppermost mantle layer parameterized by three cubic B-splines from Moho to a depth 205

of 80 km. At its base the mantle layer is continuous with an underlying layer from the 206

half-space conductive cooling model (HSCM) described in Section 3. In the inversion, 207

only four unknowns are age-dependent: sedimentary thickness and the top 3 cubic B-208

11

spline coefficients in the mantle. The other parameters are set to be constant over age. 209

Igneous crustal thickness is set constant at 7 km [e.g., White et al., 1992; Carbotte et al., 210

2008]. Crustal Vs is fixed based on an initial inversion of the 2 Ma dispersion data. 211

Fixing the igneous crust as a function of age is consistent with gravity and multichannel 212

seismic data along the ridge [Marjanovic et al., 2011] at long spatial wavelengths. The 213

Vp/Vs ratio in the igneous crust is set to be 1.76 (consistent with PREM) and is 2.0 in the 214

sediments. An additional prior constraint is imposed that the velocity gradient (dVs/dz) is 215

negative directly below Moho. In the mantle, Vp is scaled from Vs with a Vp/Vs ratio of 216

1.76 and density is scaled from Vp using results from Karato [1993]. This choice has 217

little effect on the results of the inversion. 218

2.2.2 Q model 219

Shear wave speeds in the mantle are affected both by temperature and anelasticity. The 220

inversion for a seismic model, therefore, requires the assumption of a shear Q-model. For 221

the crust we set Qμ to be consistent with PREM such that it is 80 in the sediments and 600 222

in the igneous crust. For the mantle, the principal observations of Qμ for young oceanic 223

lithosphere (near the East Pacific Rise) were obtained by Yang et al. [2007]. The center of 224

their period band is about 40 sec, where they estimated Qμ to lie between about 150 and 225

250 at depths ranging from about 10 to 40 km, with Qμ decreasing at greater depths. We 226

follow Shapiro et al. [2004] (and many others) and use a temperature and frequency 227

dependent shear Q model of the following form: 228

Q(w ) = Awa exp(a(E +PV ) / RT ) (4) 229

12

where ω is frequency in rad/sec, R is the gas constant, P is pressure, T is temperature 230

from the half-space cooling model (Fig. 5a) described later, and activation volume V = 231

1.0x10-5

m3/mol. We set α = 0.1 and activation energy E = 2.5x10

5 J/mol, which are 232

lower values than used by Shapiro et al. but more consistent with those in the study of 233

Harmon et al. [2009]. In the shallow mantle, E is larger than PV so that temperature 234

effects on Q dominate over pressure effects. Thus, what matters is the product αE, with 235

larger values accentuating the dependence on temperature. Larger values of α or E would 236

tend to raise Q more in the lithosphere relative to the underlying asthenosphere. Because 237

mantle temperatures are not well known, we choose parameters in equation (4) to make 238

the effect of temperature relatively weak. In any event, as Figure 5a shows, age-239

dependent temperature differences are important only above about 25 km depth in the 240

half-space cooling model. 241

Inserting these values into equation (4), A ≈ 30 would be consistent with Yang et al. 242

[2007] and Harmon et al. [2009], producing Qμ ≈ 175 at 30 km depth at 40 sec period. 243

With this value of A, Qμ at 10 sec period (near the center of our frequency band) is plotted 244

in Figure 5b. Three lithospheric ages are shown, using the three temperature profiles of 245

Figure 5a, which shows that temperature effects on Q are important mostly in the top 20 246

km. Below 30 km depth, Qμ is largely age-independent and equal to about 200 for A = 30. 247

It is the Q model with A = 30 that we use in producing the mantle model presented later 248

in the paper. 249

The coefficient A controls the depth-averaged Q-value in the mantle. Physically, A will 250

decrease by reducing grain size or increasing dissolved water content or retained 251

interstitial partial melt fraction [e.g., Faul et al., 2004; Faul and Jackson, 2005; Behn et al., 252

13

2009; Goes et al., 2012]. Setting A = 15 or A = 50, produces a discrete offset in Q below 253

30 km to about 100 or 350, respectively, as Figure 5b shows. The choice of A is probably 254

more important in determining the Vs model than the choice of the temperature model or 255

the other parameters in equation (4). We return later to consider the effect on the final 256

mantle Vs model of changing A from 30 to both 15 and 50 and, therefore, depth-averaged 257

Qμ from 200 to 100 and 350. 258

We present the final model at 1 sec period, extrapolating from the period band of 259

inversion using the physical dispersion correction of Minster and Anderson [1981]. 260

2.2.3 The prior distribution 261

The inversion is performed using a Bayesian Monte Carlo formalism, which has been 262

described in detail and applied systematically to EarthScope USArray data by Shen et al. 263

[2013a,b]. An input model that defines the prior distribution is initially computed by 264

performing an inversion with the dispersion curves at 2 Ma in which we allow the 265

coefficients of the crustal B-splines to vary. The igneous crust for all ages is fixed at the 266

result of this inversion. The forward problem is computed using the code of Herrmann 267

[http://www.eas.slu.edu/eqc/eqccps.html]. The best fitting model at 2 Ma (M0) is then 268

used to construct the model space for the age-dependent inversion. The model space 269

defining the prior distribution at each age is generated as follows. The sedimentary layer 270

thickness is allowed to vary ±100% relative to M0. The top first, second and third cubic 271

B-splines in the mantle are allowed to vary by ±4%, ±2% and ±1%, respectively, relative 272

to M0, which acts to squeeze heterogeneity towards shallow depth. The models at all ages 273

reach the same deep asymptotic value at 80 km depth, which is continuous with the 274

14

HSCM. Models are accepted into the posterior distribution or rejected according to the 275

square root of the reduced χ2 value. A model m is accepted if χ(m) < χmin + 0.5, where χmin 276

is the χ value of the best fitting model. After this, the mean and standard deviation of the 277

posterior distribution at each age are computed at each depth, where the mean is the 278

model we present (e.g., Fig. 6), and twice the standard deviation is interpreted as model 279

uncertainty. 280

2.2.4 Results 281

We estimate 1-D Vsv models from the mean of the posterior distribution using the 282

dispersion curves at crustal ages of 0.5, 1.0, 1.5, 2.0, 2.5, 3.0 and 3.5 Ma. The major 283

products are an age-independent igneous crust with a thickness of 7 km, a constant Vs 284

sedimentary layer with age-variable thickness, and age-dependent Vsv as a function of 285

depth in the uppermost mantle. Water depth and sedimentary thickness as a function of 286

age are presented in Figure 6a. Sediments are estimated to increase in thickness from 287

about 100 m at 0.5 Ma to about 400 m at 3.5 Ma, and the depth to the top of the igneous 288

crust increases approximately linearly with age by about 500 m between 0.5 Ma and 3.5 289

Ma. This is consistent with results from multichannel seismic (MCS) data [Carbotte et al., 290

2008]. The age-independent igneous crustal model is presented in Figure 6b. The mantle 291

age-dependent shear velocity profiles appear in Figure 6c. Shear wave speeds increase 292

with age monotonically and converge by about 60 km depth below which we have little 293

resolution. Age-dependent posterior distributions at depths of 20 km and 40 km (Fig. 6d-e) 294

illustrate the model uncertainties and show the separation of the ensemble of accepted 295

models at different ages. The posterior distributions reflect both prior information and the 296

Rayleigh wave phase velocity data, however, and their narrowness in part reflects the 297

15

tight constraints provided by the prior information. Still, the final age-dependent model 298

fits the data very well, as Figure 4 illustrates. The introduction of other variables in the 299

inversion is not justified by the need to fit the observations. 300

A low velocity zone in the uppermost mantle between 15 and 40 km depth is most 301

pronounced at young crustal ages. Unfortunately, due to a shortage of paths along the 302

ridge we are unable to provide information for lithospheric ages younger than about 0.5 303

Ma. At the youngest age (0.5 Ma) in our study, the minimum Vsv reaches ~4.07 km/sec 304

at 20 km depth. With uncertainties defined as the standard deviation of the posterior 305

distribution at each depth (e.g., Fig. 6d-e), at 20 km depth Vsv increases from 4.07 ± 0.02 306

km/sec at 0.5 Ma to 4.37 ± 0.02 km/sec at 3 Ma. At 40 km depth, Vsv increases from 4.16 307

± 0.01 km/sec at 0.5 Ma to 4.28 ± 0.01 km/sec at 3 Ma. At greater depths both the age 308

variation and uncertainties reduce because prior constraints strengthen. 309

As discussed above, the choice of the shear Q-model will affect the estimated shear 310

velocity model in the mantle. Figure 7 quantifies the effect of choosing A = 15, 30, or 50 311

in equation (4), or Q values equal to about 100, 200, or 350 below 30 km depth (with 312

somewhat higher values in the shallower mantle arising from cooler temperatures). 313

Lowering mantle Q increases Vs in the estimated model, but this range of Q models 314

produces Vs models within the model uncertainty. Thus, the choice of the Q model 315

amongst these alternatives will not affect the conclusions reached in this paper. Much 316

lower Q values at young lithospheric ages, as advocated for example by Faul and Jackson 317

[2005], would further increase Vs in the shallow mantle. If such low Q values were to 318

exist, however, they would probably result from partial melt. In section 3, we invoke the 319

existence of partial melt near the ridge in order to explain the low shallow shear wave 320

16

speeds we observe near the ridge crest. Thus, whether we explain the observations with 321

low shear wave speeds (as we prefer) or exceptionally low Q near the ridge crest, partial 322

melt would be inferred in either case. 323

3. Discussion and Conclusions 324

The age-dependent mantle Vsv model is summarized in Figure 8a, which also presents 325

the distance to the Juan de Fuca ridge (converted from age by using a half spreading rate 326

of ~30 km/Ma; [Wilson, 1993]). This 2-D plot is contoured with solid or dashed lines 327

every 0.05 km/sec with solid lines at shear wave speeds of 4.2, 4.3, 4.4, and 4.5 km/s and 328

dashed lines at 4.15, 4.25, and 4.35, and 4.45 km/s. This model is compared with shear 329

velocities converted from the thermal half-space conductively cooling model (HSCM) in 330

Figure 8b. Temperature profiles of the HSCM at several ages are plotted in Figure 5a. In 331

constructing the HSCM [Turcotte and Schubert, 2002], we use a mantle potential 332

temperature of 1315 °C and a thermal diffusivity of 10-6

m2/s, convert to anharmonic Vs 333

using the approximation of Stixrude and Lithgow-Bertelloni [2005], and model the effect 334

of anelasticity using the correction of Minster and Anderson [1981] based on the shear Q 335

model of eqn. (4) with A = 30. The Vs model from the HSCM is presented at 1 sec period 336

to match the observed model. The predicted shear wave speed from the HSCM is 337

isotropic Vs, whereas the model inferred from Rayleigh wave dispersion is Vsv. 338

Knowledge of radial anisotropy in the upper mantle would allow for a correction between 339

these values, but without Love waves we do not even know the relative sizes of Vsv and 340

Vsh. However, |Vsv-Vsh| is probably less than 3% [Ekstrom and Dziewonski, 1998], and 341

may be much smaller [e.g., Dunn and Forsyth, 2003, Harmon et al., 2009] in the shallow 342

mantle near the ridge, so the effect on Vs is almost certainly within ±1% assuming a 343

17

Voigt-average of Vsv and Vsh. If this value were constant across the study region and we 344

were to use it to convert the estimated Vsv to Vs in Figure 8a, the transformation would 345

shift the mean at each depth but not the variation with age. Thus, the estimated age 346

variation is expected to be robust relative to the introduction of radial anisotropy into the 347

model. 348

As observed in Figure 8a-b, both the estimated model and the HSCM model possess a 349

monotonically thickening high velocity lid at shallow mantle depths, and both have 350

similar average shear wave speeds in the upper mantle of ~4.25 km/sec. There are also 351

prominent differences between them. (1) First, the fast lid is observed to thicken at a 352

faster rate than for the HSCM. If we define the base of the lid (or the base of the 353

lithosphere) to be at 4.3 km/s, then by about 3.5 Ma (~100 km from the ridge) the 354

estimated lid thickens to ~40 km but the lid in the HSCM only penetrates to less than 30 355

km depth. Although the choice of 4.3 km/s is ad-hoc, the observed lithospheric lid is 356

probably more than 1.3 times thicker than predicted by the HSCM. The faster 357

development of the lithospheric lid than predicted by the HSCM may imply non-358

conductive cooling processes, such as convection or the vertical advection of fluids in the 359

shallow mantle. 360

(2) A second major difference is that the estimated model possesses a prominent low 361

velocity zone (LVZ) in the uppermost mantle (15-40 km) at young ages near the ridge 362

(age < 1.5 Ma), but such low wave speeds are not present in the HSCM. Low shear 363

velocities in the mantle (<4.1 km/sec) at 15-40 km beneath the ridge also have been seen 364

beneath the East Pacific Rise [Dunn and Forsyth, 2003; Yao et al., 2011], which was 365

attributed to partial melt beneath the ridge. 366

18

Using physically more sophisticated models than the HSCM, Goes et al. [2012] show that 367

if the upper mantle is depleted in basalt, resulting in a harzburgite composition of the 368

residue, but retains dissolved water, then Vs would be far lower than what we observe in 369

the uppermost mantle near the Juan de Fuca ridge. However, with a largely dehydrated 370

dry or merely damp depleted mantle devoid of partial melt, no LVZ appears and Vs is 371

very similar to the HSCM as can be seen in Figure 8d. The principal difference between 372

this model and the HSCM is more rapid cooling in the shallow mantle and the 373

development of a thicker lid. This difference arises principally because Goes et al. 374

include the effects of convection. They also use a more sophisticated PT-velocity 375

conversion, which may also have contributed to the difference. 376

In contrast Goes et al. have also included a retained partial melt fraction with a maximum 377

of about 1%. Using the Qg model defined in their paper, they produce the Vs model 378

shown in Figure 8c, which displays a shallow low velocity zone between 10 and 50 km 379

depth that is qualitatively similar to our model but with minimum shear velocities that are 380

lower and with low shear velocities extending farther from the ridge. However, they take 381

their partial derivatives of anharmonic Vs relative to a melt fraction from the highest 382

values of Hammond and Humphreys [2000] and, therefore, may have over-predicted the 383

effect of partial melt on Vs. Still, our results are probably consistent with a retained melt 384

fraction somewhat smaller than 1%, although this value is very poorly determined. 385

These observations lead us to conclude that the low shear wave speeds that we observe 386

near the Juan de Fuca Ridge probably derive from a small retained melt fraction less than 387

about 1% in a largely dry depleted harzburgitic uppermost mantle. In addition, the 388

amplitude of the observed LVZ diminishes with age, which is consistent with cooling and 389

19

the reduction in the melt fraction. By 1.0-1.5 Ma, the velocity minimum at about 20 km 390

has largely disappeared, which, following the interpretation presented here, would 391

probably mean that partial melt is largely absent past about 1.0 Ma (i.e., 30 km from the 392

ridge crest). 393

This study was performed with only six months of OBS data acquired near the Juan de 394

Fuca ridge. Since the study’s completion, longer time series have been accruing and other 395

data have become available including higher sampling rates, horizontal components, and 396

stations nearer to the continent. Further analysis of these data as well as the assimilation 397

of other types of data (e.g., receiver functions, heat flow measurements, etc.) are expected 398

to extend the present study considerably. 399

400

20

Acknowledgments. The authors are grateful to the two anonymous reviewers for 401

constructive criticisms that improved this paper and to Saskia Goes for providing the data 402

used to construct Figs. 8c-d, and for self-consistently computing these values at 1 sec 403

period for comparison with our results. They are also particularly grateful to the Cascadia 404

Initiative Expedition Team for acquiring the Amphibious Array Ocean Bottom 405

Seismograph data and appreciate the open data policy that made the data available shortly 406

after they were acquired. The facilities of the IRIS Data Management System were used 407

to access some of the data used in this study. The IRIS DMS is funded through the US 408

National Science Foundation under Cooperative Agreement EAR-0552316. 409

410

21

References 411

Amante, C. and B. W. Eakins (2009), ETOPO1 1 Arc-Minute Global Relief Model: 412

Procedures, Data Sources and Analysis. NOAA Technical Memorandum NESDIS 413

NGDC-24, 19 pp. 414

Barmin, M. P., M. H. Ritzwoller, and A. L. Levshin (2001), A fast and reliable method for 415

surface wave tomography, Pure and Applied Geophysics, 158(8), 1351–1375, 416

doi:10.1007/PL00001225. 417

Behn, M.D., G. Hirth, and J.R. Elsenbeck II (2009), Implications of grain size evolution 418

on the seismic structure of the oceanic upper mantle, Earth Planet. Sci. Lett., 282, 419

178-189, doi:10.1016/j.epsl.2009.03.014. 420

Bensen, G.D., M.H. Ritzwoller, M.P. Barmin, A.L. Levshin, F. Lin, M.P. Moschetti, N.M. 421

Shapiro, and Y. Yang, Processing seismic ambient noise data to obtain reliable 422

broad-band surface wave dispersion measurements, Geophys. J. Int., 169, 1239-423

1260, doi: 10.1111/j.1365-246X.2007.03374.x, 2007. 424

Carbotte, S. M., M. R. Nedimović, J. P. Canales, G. M. Kent, A. J. Harding, and M. 425

Marjanović (2008), Variable crustal structure along the Juan de Fuca Ridge: 426

Influence of on-axis hot spots and absolute plate motions, Geochem. Geophys. 427

Geosyst., 9, Q08001, doi:10.1029/2007GC001922. 428

Dunn, R. A., and D. W. Forsyth (2003), Imaging the transition between the region of 429

mantle melt generation and the crustal magma chamber beneath the southern East 430

Pacific Rise with short-period Love waves, J. Geophys. Res., 108(B7), 2352, 431

doi:10.1029/2002JB002217. 432

22

Ekstrom, G., and A. M. Dziewonski (1998), The unique anisotropy of the Pacific upper 433

mantle, Nature, 394(6689), 168–172, doi:10.1038/28148. 434

Faul, U.H., J.D. Fitz Gerald, and I. Jackson (2004), Shear wave attenuation and 435

dispersion in melt-bearing olivine polycrystals: 2. Microstructural interpretation 436

and seismological implications, J. Geophys. Res., B06202, 437

doi:10.1029/2003JB002407. 438

Faul, U.H., and I. Jackson (2005), The seismological signature of temperature and grain 439

size variations in the upper mantle, Earth Planet. Sci. Lett., 234(1-2), 119-134, 440

doi:10.1016/j.epsl.2005.02.008. 441

Goes, S., J. Armitrage, N. Harmon, H. Smith, and R. Huismans (2012), Low seismic 442

velocities below mid-ocean ridges: Attenuation versus melt retention, J. Geophys. 443

Res., 117, B12403, doi:10.1029/2012JB009637. 444

Hammond, W. C., and E. D. Humphreys (2000), Upper mantle seismic wave velocity: 445

Effects of realistic partial melt geometries, J. Geophys. Res., 105, 10,975–10,986. 446

Harmon, N., D. Forsyth, and S. Webb (2007), Using ambient seismic noise to determine 447

short-period phase velocities and shallow shear velocities in young oceanic 448

lithosphere, Bulletin of the Seismological Society of America, 97(6), 2009–2023, 449

doi:10.1785/0120070050. 450

Harmon, N., D. W. Forsyth, and D. S. Weeraratne (2009), Thickening of young Pacific 451

lithosphere from high-resolution Rayleigh wave tomography: A test of the 452

conductive cooling model, Earth and Planetary Science Letters, 278, 96–106, 453

doi:10.1016/j.epsl.2008.11.025. 454

23

Karato, S. (1993), Importance of anelasticity in the interpretation of seismic tomography, 455

Geophys. Res. Lett., 20(15), 1623–1626, doi:10.1029/93GL01767. 456

Levshin, A.L. and M.H. Ritzwoller, Automated detection, extraction, and measurement of 457

regional surface waves, Pure Appl Geophys, 158(8), 1531 - 1545, 2001. 458

Lin, F., M.H. Ritzwoller, J. Townend, M. Savage, S. Bannister, Ambient noise Rayleigh 459

wave tomography of New Zealand, Geophys. J. Int., 18 pages, 460

doi:10.1111/j.1365-246X.2007.03414.x, 2007. 461

Lin, F.-C., M. H. Ritzwoller, and R. Snieder (2009), Eikonal tomography: surface wave 462

tomography by phase front tracking across a regional broad-band seismic array, 463

Geophysical Journal International, 177(3), 1091–1110, doi:10.1111/j.1365-464

246X.2009.04105.x. 465

MELT Seismic Team, The (1998), Imaging the deep seismic structure beneath a mid-466

ocean ridge: The MELT experiment, Science, 280, 1215–1218, 467

doi:10.1126/science.280.5367.1215. 468

Marjanović, M., S. M. Carbotte, M. R. Nedimović, and J. P. Canales (2011), Gravity and 469

seismic study of crustal structure along the Juan de Fuca Ridge axis and across 470

pseudofaults on the ridge flanks, Geochem. Geophys. Geosyst., 12, Q05008, 471

doi:10.1029/2010GC003439. 472

Minster, J.B. and D.L. Anderson (1981), A model of dislocaiton-controlled rheology for 473

the mantle, Phil. Trans. R. Soc. London, 299, 319-356. 474

Mueller, R. D., W. R. Roest, J.-Y. Royer, L. M. Gahagan, and J. G. Sclater (1997), Digital 475

isochrons of the world’s ocean floor, J. Geophys. Res., 102, 3211 –3214. 476

24

Shapiro, N.M. and M.H. Ritzwoller, Thermodynamic constraints on seismic inversions, 477

Geophys. J. Int., 157, 1175-1188, doi:10.1111/j.1365-246X.2004.02254.x, 2004. 478

Shen, W., M.H. Ritzwoller, V. Schulte-Pelkum, F.-C. Lin, Joint inversion of surface wave 479

dispersion and receiver functions: A Bayesian Monte-Carlo approach, Geophys. J. 480

Int., 192, 807-836, doi:10.1093/gji/ggs050, 2013a. 481

Shen, W., M.H. Ritzwoller, and V. Schulte-Pelkum, A 3-D model of the crust and 482

uppermost mantle beneath the central and western US by joint inversion of 483

receiver functions and surface wave dispersion, J. Geophys. Res., 484

doi:10.1029/2012JB009602, 118, 1-15, 2013b. 485

Stehly, L., M. Campillo, and N.M. Shapiro, Travel time measurements from noise 486

correlations: stability and detection of instrumental errors, Geophys. J. Int., 171, 487

223–230, doi: 10.1111/j.1365-246X.2007.03492.x, 2007 488

Stixrude, L., and C. Lithgow-Bertelloni (2005), Mineralogy and elasticity of the oceanic 489

upper mantle: Origin of the low-velocity zone, J. Geophys. Res., 110, B03204, 490

doi:10.1029/2004JB002965. 491

Sun, Y.F., 2000, Core-log-seismic integration in hemipelagic marine sediments on the 492

eastern flank of the Juan de Fuca Ridge, in Fisher, A., Davis, E., and Escutia, C. 493

(Eds.), ODP Scientific Results, 168, 21-35. 494

Turcotte, D.L. and G. Schubert, Geodynamics, Cambridge University Press, 2nd

ed., 472 495

pp., 2002. 496

25

White, R. S., D. McKenzie, and R. K. O'Nions (1992), Oceanic crustal thickness from 497

seismic measurements and rare earth element inversions, J. Geophys. Res., 498

97(B13), 19683–19715, doi:10.1029/92JB01749. 499

Wilson, D. S. (1993), Confidence intervals for motion and deformation of the Juan de 500

Fuca Plate, J. Geophys. Res., 98(B9), 16053–16071, doi:10.1029/93JB01227. 501

Yang, Y, D. Forsyth, and D.S. Weeraratne (2007), Seismic attenuation near the East 502

Pacific Rise and the origin of the low-velocity zone, Earth Planet. Sci. Lett., 258, 503

260-268, doi:10.10106/j.epsl.2007.03.040. 504

Yao, H., P. Gouédard, J. A. Collins, J. J. McGuire, and R. D. van der Hilst (2011), 505

Structure of young East Pacific Rise lithosphere from ambient noise correlation 506

analysis of fundamental- and higher-mode Scholte-Rayleigh waves, Comptes 507

Rendus Geoscience, 343(8–9), 571–583, doi:10.1016/j.crte.2011.04.004. 508

509

26

Figure Captions 510

Figure 1. (a) The locations of the 18 long period Cascadia Initiative OBS stations used in 511

this study (triangles) are plotted over bathymetry with the Juan de Fuca Ridge shown as 512

the grey line. The red star (denoted CHS) marks the approximate location of the Cobb hot 513

spot. (b) The 106 inter-station ray paths are plotted with grey lines over lithospheric age 514

[Mueller et al., 1997]. (c) Example 6 month cross-correlation for data from stations J29A 515

and J47A, marked as red triangles bounding the red inter-station path in (b). The 516

waveform is colored red or blue for the positive or negative correlation lag with group 517

speeds corresponding to the fundamental mode. (d) Rayleigh wave velocity versus 518

period (FTAN) diagram of the symmetric component of the signal shown in (c). 519

Background color indicates the spectral amplitude and group and phase speeds are shown 520

with red and white circles, respectively. 521

Figure 2. (a) Estimate of measurement error. The histogram shows the distribution of the 522

differences between measurements of positive and negative lag inter-station phase times 523

for the 14 sec Rayleigh wave. (mean = -0.057 sec, st dev = 0.77 sec is taken as 524

measurement error). (b) Non-detection of a timing error. Each dot is the mean time 525

difference for a particular station between the positive and negative lags (associated with 526

outgoing and incoming waves) for the 14 sec Rayleigh wave. Red dashed and solid lines 527

indicate the 1-σ and 2-σ confidence intervals, respectively. No mean difference is outside 528

of the 2-σ confidence interval. 529

Figure 3. (a) Solid lines are the estimated age-dependent Rayleigh wave phase velocities 530

(eqn. (1)) at 7 (red), 8 (orange), 10 (green) and 15 (blue) sec period. Colored dots are the 531

27

measured inter-station phase velocities plotted at the average of the lithospheric age along 532

the inter-station path. (b) The same as (a), but for Rayleigh wave group velocities at the 533

same periods. (c)-(d) Blue histograms are the misfit (in percent) to the observed inter-534

station phase velocities at (c) 7 sec and (d) 15 sec period produced by the estimated age-535

dependent phase speed curves (eqn (1)). At 7 sec and 15 sec period, respectively, mean 536

misfits are 0.1% and 0.02% and the standard deviations of the misfits are 1.8% and 0.9%. 537

The red histograms are the misfits based on the 0.5 Ma model. At 7 and 15 sec period, 538

respectively, mean misfits from this age-independent model are -9.7% and 3.2% and the 539

standard deviations of the misfits are 5.7% and 1.4%. Thus, the age-dependent model 540

significantly reduces the standard deviation of the misfit compared with an age-541

independent model and produces a nearly zero-mean misfit. 542

Figure 4. Estimated dispersion curves for seafloor ages of 1 Ma (red) and 3 Ma (black). 543

Error bars are the measured Rayleigh wave phase velocity and the estimated 1 standard 544

deviation uncertainty. Solid curves are the predictions from the inverted age-dependent 545

shear velocity model (Fig. 6). 546

Figure 5. (a) Examples of the mantle temperatures from the half-space conductive 547

cooling model (HSCM) plotted for three lithospheric ages. This temperature model is 548

used in the Q-model (eqn. (4)). (b) Examples of Qμ for three different lithospheric ages 549

for three different values of the A coefficient of equation (4). 550

Figure 6. Estimated model. (a) Water depth (blue line), estimated sedimentary layer 551

thickness (red line), and the estimated depth of crystalline basement below the ocean 552

surface (grey line), which is the sum of water depth and sedimentary layer thickness. (b) 553

28

Estimated crustal Vs model, which varies in age only by sediment thickness and water 554

depth. (c) The estimated age-dependent shear-wave velocity models (Vsv) in the mantle 555

from 0.5 to 3 Ma. The mean of the estimated posterior distribution is shown for each age. 556

The age legend at lower left corresponds both to (b) and (c). (d) Posterior distributions of 557

Vsv models for each seafloor age at 20 km depth. (e) Same as in (d), but for 40 km depth. 558

All models are presented at 1 sec period. 559

Figure 7. Effect of varying the Q model on estimates of Vs in the mantle. Vs models 560

determined using three different Q models with varying A values (eqn. (4)) are shown: A 561

= 15 (red), 30 (black), 50 (blue). We use A = 30 in this paper, and the estimated 2 562

standard deviation uncertainty in the resulting model is shown with the grey corridor. 563

Figure 8. Comparison of (a) our estimated Vsv model and (b) the half-space conductive 564

cooling model (HSCM) as a function of seafloor age. For further comparison, models 565

from Goes et al. [2012] are presented in (c) and (d) with and without retained melt, 566

respectively. Shear wave speeds in increments of 0.1 km/sec are contoured with solid 567

lines and values in odd multiples of 0.05 km/sec are contoured with dashed lines. All 568

models are converted to 1 sec period for comparison. (S. Goes provided the models in (c) 569

and (d) and converted them to 1 sec period self-consistently.) 570

Study Area

130˚W 128˚W 126˚W

44˚N

45˚N

46˚N

47˚N

48˚N

49˚N

0 1000 2000 3000 4000 5000Bathymetry (m)

J23AJ28A

J30AJ31A

J37AJ38A

J39A

J45AJ46A

J52AJ53A

J54AJ55A

J63A J67A

J68A

J29A

J47A

CHS

(a) Path Coverage

130˚W 128˚W 126˚W

44˚N

45˚N

46˚N

47˚N

48˚N

49˚N

0.0 1.5 3.0 4.5 6.0 7.5 9.0Lithospheric Age (Ma)

J29A

J47A

(b)

0

−400 −200 0 200 400Time (sec)

Time Series(c)

0.5

1.5

2.5

3.5

4.5

Vel

ocity

(km

/sec

)

6 7 8 9 10 20Period (sec)

FTAN Diagram(d)

0

5

10

15

20

25

30

35

No.

of I

nter

−st

atio

n P

aths

−5 0 5time (sec)

Phase Time Misfit

mean −0.057 std 0.77

(a)

2

4

6

8

10

12

14

No.

of M

easu

rem

ents

Per

Sta

tion

−1.0 −0.5 0.0 0.5 1.0time (sec)

Station Mean Misfits1−σ2−σ

(b)

2.5

3.0

3.5

4.0

Vel

ocity

(km

/sec

)

1 2 3 4Age (Ma)

Phase−Age Relationships(a)

15 sec10 sec

8 sec7 sec 0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Vel

ocity

(km

/sec

)

1 2 3 4Age (Ma)

Group−Age Relationships(b)

15 sec10 sec

8 sec7 sec

0

10

20

30

40

50

Per

cent

age

(%)

−30 −20 −10 0 10Phase Vel Misfit (%)

Misfit 7.0 sec

age−dep: mean 0.10% std 1.8%0.5 Ma: mean −9.7% std 5.7%

(c)

0

10

20

30

40

50

Per

cent

age

(%)

−5 0 5 10Phase Vel Misfit (%)

Misfit 15.0 sec

age−dep: mean 0.018% std 0.85%0.5 Ma: mean 3.2% std 1.4%

(d)

1.0

1.5

2.0

2.5

3.0

3.5

4.0

6 8 10 12 14 16 18 20Period (sec)

Velo

city

(km

/sec

)

1 Ma3 Ma

10

20

30

40

50

60

10

20

30

40

50

60100 200 300 400 500 600

100 200 300 400 500 60010

20

30

40

50

601200 1350 1500 1650

1200 1350 1500 1650

3 Ma

0.5 Ma

1.5 Ma

Q

Age/A dependent Q Temperature

Temperature (K)

Dep

th (k

m)

3 Ma

0.5 Ma

1.5 Ma

(a) (b)

A = 30A = 15 A = 50

Age (Ma)

Thic

knes

s (km

)

0.00.10.20.30.40.5

2.4

2.6

2.8

3.0

3.2

0.5 1.0 1.5 2.0 2.5 3.0 3.5

2468

103.2 3.4 3.6 3.8 4.0

Dep

th (k

m)

Vs (km/sec)

Dep

th (k

m)

Vs (km/sec)

0.5 Ma1 Ma1.5 Ma

2 Ma2.5 Ma3 Ma

Dep

th (k

m)

(a)

(b)

(c)

water depthsediment thickness

10

20

30

40

504.0 4.1 4.2 4.3 4.4 4.5

0

10

20

0

10

20

4.0 4.1 4.2 4.3 4.4

0

10

20

30

0

10

20

30

4.10 4.15 4.20 4.25 4.30Vs (km/sec)

Vs (km/sec)

Perc

enta

ge (%

)Pe

rcen

tage

(%)

0.5 Ma 1 Ma 2 Ma 3 Ma

Vs posterior distributions

Depth = 20 km

Depth = 40 km

(d)

(e)

10

20

30

40

504.0 4.1 4.2 4.3 4.4 4.5

A = 50

A = 15

A = 300.5 Ma 2 Ma

Dep

th (k

m)

Vs (km/sec)

20

30

40

500.5 1.0 1.5 2.0 2.5 3.0 3.5

15 30 45 60 75 90 105

4.10 4.15 4.20 4.25 4.30 4.35 4.40 4.45 4.50

20

30

40

500.5 1.0 1.5 2.0 2.5 3.0 3.5

15 30 45 60 75 90 105

20

30

40

500.5 1.0 1.5 2.0 2.5 3.0 3.5

15 30 45 60 75 90 105

20

30

40

500.5 1.0 1.5 2.0 2.5 3.0 3.5

15 30 45 60 75 90 105

Estimated Vs model HSCM

Qg, damp, 1% melt Qg, damp, no melt

Vs (km/sec)

Dep

th (k

m)

Dep

th (k

m)

Dep

th (k

m)

Dep

th (k

m)

Age (Ma) Age (Ma)

Age (Ma) Age (Ma)

Distance to Ridge (km)

Distance to Ridge (km)

(a) (b)

(c) (d)Distance to Ridge (km)

Distance to Ridge (km)